Transcriptional profiles of supragranular-enrichedgenes associate with corticocortical networkarchitecture in the human brainFenna M. Krienena,1,2, B. T. Thomas Yeob,c, Tian Gec,d, Randy L. Bucknerc,e,f, and Chet C. Sherwooda

aDepartment of Anthropology, Center for the Advanced Study of Human Paleobiology and Institute for Neuroscience, The George Washington University,Washington, DC 20052; bDepartment of Electrical and Computer Engineering, Clinical Imaging Research Centre, Singapore Institute for Neurotechnology& Memory Networks Program, National University of Singapore, Singapore 117583; cAthinoula A. Martinos Center for Biomedical Imaging, MassachusettsGeneral Hospital, Charlestown, MA 02129; dPsychiatric and Neurodevelopmental Genetics Unit, Center for Human Genetic Research, Massachusetts GeneralHospital, Boston, MA 02114; eDepartment of Psychiatry, Massachusetts General Hospital, Boston, MA 02114; and fDepartment of Psychology and Center forBrain Science, Harvard University, Cambridge, MA 02138

Edited by Robert Desimone, Massachusetts Institute of Technology, Cambridge, MA, and approved December 10, 2015 (received for review June 3, 2015)

The human brain is patterned with disproportionately large, distrib-uted cerebral networks that connect multiple association zones in thefrontal, temporal, and parietal lobes. The expansion of the corticalsurface, along with the emergence of long-range connectivitynetworks, may be reflected in changes to the underlying moleculararchitecture. Using the Allen Institute’s human brain transcrip-tional atlas, we demonstrate that genes particularly enriched insupragranular layers of the human cerebral cortex relative tomouse distinguish major cortical classes. The topography of tran-scriptional expression reflects large-scale brain network organiza-tion consistent with estimates from functional connectivity MRIand anatomical tracing in nonhuman primates. Microarray expres-sion data for genes preferentially expressed in human upper layers(II/III), but enriched only in lower layers (V/VI) of mouse, werecross-correlated to identify molecular profiles across the cerebralcortex of postmortem human brains (n = 6). Unimodal sensory andmotor zones have similar molecular profiles, despite being distrib-uted across the cortical mantle. Sensory/motor profiles were anti-correlated with paralimbic and certain distributed association networkprofiles. Tests of alternative gene sets did not consistently distinguishsensory and motor regions from paralimbic and association re-gions: (i) genes enriched in supragranular layers in both humansand mice, (ii) genes cortically enriched in humans relative to non-human primates, (iii ) genes related to connectivity in rodents,(iv) genes associated with human and mouse connectivity, and(v) 1,454 gene sets curated from known gene ontologies. Molecularinnovations of upper cortical layers may be an important compo-nent in the evolution of long-range corticocortical projections.

corticocortical connectivity | human transcriptome | association cortex |supragranular | brain evolution

Patterns of gene expression in the cerebral cortex are generallyconserved across species, reflecting strong constraints in the

development and evolution of cortical architecture (1–6). Pre-vious work examining transcriptional variation in nonhumanprimates and rodents indicate that molecular similarities be-tween cortical regions in the adult brain are best explained byspatial proximity (7, 8). Molecular variation often takes the formof graded expression along a principal axis, in many cases appearingas rostrocaudal gradients across the cortex (8). The strong tendencyfor transcriptional variation to follow spatial proximity in the adultcortex likely reflects both functional gradients, as well as their de-velopmental origins in terms of physical and temporal adjacencyduring neurogenesis of cells destined for neighboring locations inthe cortex (at least for cells derived from the ventricular prolifera-tive pool) (7).Spatial proximity likely captures the major features governing

how molecular profiles vary across the cortex in all species, in-cluding humans. However, the expansion of the cerebral cortexin primates generally, and humans specifically, was accompanied

by changes to both microstructural and connectional organization(9–11). In particular, connectivity patterns in humans, mappedby noninvasive imaging techniques, reveal a set of distributed,interdigitated networks that tile the expanded portions of theassociation cortex. These distributed networks have certain or-ganizational properties that depart from evolutionarily con-served unimodal sensory and motor zones, where connectionaltopography between areas or fields is often densest betweenneighboring locations (12). Higher-order cortical regions alsopossess local connections, but are distinguished by the relativeprevalence of long-range connections (13–15). An open questionis how the molecular architecture underlying these differentcortical classes supports their distinct connectivity patterns. Inparticular, the distributed nature of networks that interconnectthe prefrontal, parietal, temporal, and cingulate association cortextogether raises the possibility that innovations in molecular pro-files will be associated with the emergence of extended forms oflong-range connectivity in those regions (16, 17).A recent study of expression profiles of 1,000 genes in the

mammalian cerebral cortex found that 79% have conserved lam-inar expression patterns between mice and humans (18). Severalof the remaining 21% exhibited species- or region-specific dis-tributions. Some had different laminar expression patterns in

Significance

The human cerebral cortex is patterned with distributed net-works that connect disproportionately enlarged associationzones across the frontal, temporal, and parietal lobes. Weasked herein whether the expansion of the cortical surface,with the concomitant emergence of long-range connectivitynetworks, might be accompanied by changes to the underlyingmolecular architecture. We focused on the supragranular layersof the cortex, where most corticocortical connections originate.Genes that are enriched in supragranular layers in the humancerebral cortex relative to mouse are expressed in a topogra-phy that reflects broad cortical classes (sensory/motor, para-limbic, associational) and their associated network properties.Molecular innovations of upper cortical layers may be an im-portant component in the evolution of increased long-rangecorticocortical projections.

Author contributions: F.M.K., B.T.T.Y., T.G., R.L.B., and C.C.S. designed research; F.M.K.performed research; F.M.K., B.T.T.Y., and T.G. contributed new reagents/analytic tools;F.M.K. analyzed data; and F.M.K., B.T.T.Y., T.G., R.L.B., and C.C.S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1Present address: Department of Genetics, Harvard Medical School, Cambridge, MA 02115.2To whom correspondence should be addressed. Email: fenna_krienen@hms.harvard.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1510903113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1510903113 PNAS Early Edition | 1 of 10

NEU

ROSC

IENCE

PNASPL

US

putatively homologous brain regions across mice and humans. Inparticular, 19 genes were uniquely or disproportionately enrichedin the supragranular layers of the cerebral cortex in humans, butnot in mice. Zeng et al. (18) hypothesized that the selective en-richment of these genes may be a molecular signature of the en-hancement of long-range corticocortical projections emanatingfrom layer III pyramidal neurons.This idea is intriguing because supragranular pyramidal neu-

rons are overrepresented in primates, particularly in humans(19). Within the cortical column, pyramidal projection neuronsin layers II and III are responsible for the majority of local andlong-range intracortical connections. Several observations in-dicate that upper layers have undergone expansion and remod-eling in primates relative to rodents. Layers II and III arerelatively thicker in primates (19, 20), and layer III neurons inprimates, particularly in higher-order association cortices, haveprotracted spinogenesis (21–23). The relationship between thesesupragranular molecular innovations and corticocortical con-nectivity is still unclear. One possibility is that molecular inno-vations in upper cortical layers mediate changes to corticocorticalconnectivity in humans, particularly with respect to long-distanceprojections.To explore this possibility, the present study examines co-

variation of molecular profiles of the genes identified by Zenget al. (18) in relation to macroscale network architecture gleanedfrom functional connectivity MRI (fcMRI) estimates, and cor-roborated by tract tracing studies in nonhuman primates. Wealso tested for an association with additional gene sets, includingthose that are expressed in supragranular layers (layers II andIII) in both humans and mice (18), those that are selectivelycortically enriched in humans relative to nonhuman primates (2),those that have been previously associated with connectivity inrodents (24), and a collection of genes whose coexpression inhumans is consistent with structural connectivity in mouse (25).Additionally, 1,454 annotated gene sets from the MolecularSignatures Database (MSigDB) were used to compare patternsof covariation against a large number of biologically meaningfulgene sets.We found that transcriptional variation of genes that are se-

lectively enriched in human upper cortical layers are associatedwith major functional cortical classes (sensory/motor, paralimbic,or associational) and corresponding network topography. Othergene sets were less successful in distinguishing cortical classesand networks. Furthermore, although spatial proximity was afactor in accounting for relative variation in the expressionprofiles of the genes examined in this study, network identity wassignificantly associated with profile similarity of distributed re-gions across the cortex. The results suggest that molecular in-novations of upper layer cortical architecture may be an importantcomponent in the evolution of long-range corticocortical projec-tions in humans.

ResultsCorticocortical Connectivity Networks Have Distinct Molecular Profilesof Human Supragranular Enriched Genes. The primary gene set[Human Supragranular Enriched (HSE) set, n = 19 genes](Table 1) was selected based on in situ hybridization imagesavailable from the Allen Institute Human Transcriptional Atlas,as well as laminar characterizations in Zeng et al. (18). Genes inthis set included those whose laminar expression in the cerebralcortex is either (i) unique to humans relative to mouse, andexpressed in upper layers of the cortex, or (ii) predominantlyenriched in layer V or layers V/VI in mouse, but shift to pre-dominately layer III or layers II/III in humans.We used previously published fcMRI parcellations to explore

how transcriptional profiles vary within and across human cor-ticocortical networks. fcMRI has proven to be a powerful, if in-direct, method for estimating network topography in the human

brain (26, 27). Connectivity estimates using fcMRI are broadlyconsistent with known anatomical systems as estimated fromnonhuman primate anatomical tracing, with some exceptions (28–33). A parcellation of 17 fcMRI networks based on resting-statedata analyzed in Yeo et al. (31) was split into sets of componentsaccording to their locations and extent on the cortical surface,resulting in a set of 114 cortical regions composed of roughlysymmetric territories in the left and right hemispheres.Our analyses revealed that distinct molecular profiles of the

HSE set define different classes of cerebral cortex. These dis-sociations are consistent with groupings of large-scale cortico-cortical networks. Fig. 1A shows the correlation matrix for theHSE genes from microarray data available from six humanpostmortem brains released by the Allen Institute (human.brain-map.org/). The correlation matrix reflects the similarity of tran-scriptional profiles of brain regions distributed across the cortex.For each postmortem brain, atlas coordinates of cerebral sam-ples [based on the Montreal Neurological Institute (MNI) atlas]were used to assign each to one of the 114 17-network components(see Methods and Fig. S1 for details on network assignments andnaming conventions). Matrix rows and columns correspond to the114 regions. Fig. 1B shows the resting-state fcMRI correlationmatrix for the same 114 components (27) for comparison. Thetranscriptional profiles in Fig. 1A fall into clear groupings: regionswithin auditory, somato/motor, and visual networks tend to havesimilar molecular profiles, which are distinct from regions withinparalimbic and association networks. Primary motor and somato-sensory samples falling on the precentral and postcentral gyrus,respectively, had highly similar transcriptional expression of thegenes examined here, consistent with previous reports (34), and aregrouped together for the remainder of the paper. Regions withinparalimbic and certain association networks tend to also havesimilar molecular profiles to each other.As an illustration of transcriptional similarity across spatially

distributed locations, brain samples falling in the superior tem-poral gyrus at or near the primary auditory cortex, and belt/parabelt auditory association regions have similar transcriptionalprofiles to nearby somato/motor network regions, as well as to

Table 1. HSE gene set

Gene symbol Entrez ID

BEND5 BEN domain containing 5C1QL2 Complement component 1, q

subcomponent-like 2CACNA1E Calcium channel, voltage-dependent,

R type, α 1E subunitCOL24A1 Collagen, type XXIV, α 1COL6A1 Collagen, type VI, α 1CRYM Crystallin, μKCNC3 Potassium voltage-gated channel,

Shaw-related subfamily, member 3KCNH4 Potassium voltage-gated channel, subfamily

H (eag-related), member 4LGALS1 Lectin, galactoside-binding, soluble, 1MFGE8 Milk fat globule-EGF factor 8 proteinNEFH Neurofilament, heavy polypeptidePRSS12 Protease, serine, 12 (neurotrypsin, motopsin)SCN3B Sodium channel, voltage-gated, type III, β subunitSCN4B Sodium channel, voltage-gated, type IV, β subunitSNCG Synuclein, γ (breast cancer-specific protein 1)SV2C Synaptic vesicle glycoprotein 2CSYT2 Synaptotagmin IITPBG Trophoblast glycoproteinVAMP1 Vesicle-associated membrane protein

1 (synaptobrevin 1)

2 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1510903113 Krienen et al.

more distant occipital regions (Fig. 1C), relative to neighboringregions in the temporoparietal and middle temporal gyrus thatbelong to higher-order association networks. As another exam-ple of similar transcriptional profiles spanning long distances,regions in the default network across the prefrontal, parietal,temporal, and cingulate cortices have higher similarity to eachother than to association regions that fall outside of that network(Fig. S2). These examples highlight that transcriptional similarityoccurs not only between regions of the same cortical class (e.g.,primary sensory/sensorimotor regions), but also between regionsthat connect together to form distributed anatomical systems.Similarity of transcriptional profiles was assessed by testing

how many pairs of networks (17 × 16/2 = 136) were significantlypositively or negatively correlated. For the HSE set, 103 of 136network pairs have significant gene expression profile correla-tions [either positively or negatively correlated; q < 0.05, false-discovery rate (FDR) corrected] (Fig. S3A).

Comparison with Other Gene Sets. Multiple comparison gene setswere analyzed. Some comparison sets share relevant propertiesin common with the HSE set: (i) genes that have conservedexpression in supragranular layers in both mouse and human(Conserved Supragranular set, n = 14 genes) (18), (ii) genespreviously shown to be cortically enriched relative to subcorticalstructures in humans relative to macaque monkeys and chim-panzees (Human Cortically Enriched set, n = 20 genes) (2), and(iii) genes that were previously associated with anatomic con-nectivity in rodents and could be matched to homologous genesin the microarray dataset (Rodent Connectivity set, n = 381genes) (24). The Conserved Supragranular set contains genesthat are enriched in layers II and III in both mouse and human(18). Genes in the Human Cortically Enriched set are selectivelyup-regulated in human cerebral cortex relative to chimpanzeeand macaque cerebral cortex (2), and therefore also have human-

specific enrichment in the cerebral cortex. The Rodent Connec-tivity set contains a large number of genes previously shown topredict anatomical connectivity in rodents (24). A fourth com-parison set consisted of genes whose expression was shown to bedistinctly associated with four human cortical networks (Human/Mouse Connectivity set, n = 136) (25). The expression pattern ofthis gene set also follows structural connectivity in homologousregions in the mouse (25).Permutation testing was used to determine whether correla-

tions of transcriptional profiles are stronger for the HSE set thanfor these alternatives. Seventy-six of the 153 edges in the matrixwere significantly stronger for the HSE set relative to the Con-served Supragranular set, 77 of 153 for the HSE vs. HumanCortically Enriched set, and 103 of 153 for the Rodent Connec-tivity set (q < 0.05, FDR-corrected) (Fig. S3B). For the reversecontrasts, 15 of 153, 16 of 153, and 1 of 153 pairs were significant,respectively (q < 0.05, FDR-corrected). Twenty of 153 networkpairs were significantly stronger in the HSE vs. Human/MouseConnectivity set (q < 0.05, FDR-corrected; 0 of 153 werestronger in the reverse contrast). The HSE and Human/MouseConnectivity set share four genes in common. Because thisoverlap comprises 21% of the HSE set, it is not surprising thatsimilar correlation structure exists between these two sets. TheHuman/Mouse Connectivity set additionally shares two genes incommon with the Conserved Supragranular Enriched set andfour genes in common with the Rodent Connectivity set.An open question is whether the within- and across-network

transcriptional coexpression observed with the HSE set is presentin other sets comprising a wider ontology. A collection of 1,454gene sets obtained from the Broad Institute’s MSigDB was usedto test this possibility.We used the widely used network-based statistic (NBS) 35) to

compare the HSE set against all 1,458 alternatives (Methods).NBS reflects the extent to which connected components (sets of

A B

C

Fig. 1. Transcriptional profiles follow cortical subtypes and network topography. (A) Correlations between pairs of 114 brain regions in terms of theirtranscriptional profiles for the HSE genes. Correlations are averaged across individuals (n = 6). Complete black rows are brain regions that were not sampledby any individual. (B) Correlation matrix showing coupling patterns across the same 114 regions, measured by fcMRI at rest [adapted from Buckner et al. (27)].Regions in both matrices are arranged such that those that belong to the same fcMRI network are grouped together. (C, Left) Surface representation of the17-network parcellation used to group brain regions in A and B [adapted from Yeo et al. (31)]. White asterisks are locations of regions shown in the polar plotshown on the right. (Right) Polar plot showing transcriptional profile correlations between the auditory cortex region and surrounding regions, as well asdistant occipital regions. Note the higher similarity of transcriptional profiles between somato/motor regions and occipital visual regions than to neighboringassociation cortical regions.

Krienen et al. PNAS Early Edition | 3 of 10

NEU

ROSC

IENCE

PNASPL

US

nodes connected by suprathreshold links) are present in a graph.NBS is particularly suited for determining whether the graphstructure of two groups (here, two gene sets) differ in a coherentmanner, which is not reflected in univariate testing of each linkindependently. Across a range of tested correlation thresholds(0.1–0.4), the HSE maximal connected component size was sig-nificantly larger than that of alternatives (all P < 0.05), with theexception of the Human/Mouse Connectivity set, in which themaximal component size of the HSE set was larger but the dif-ference did not reach statistical significance (P = 0.07–0.12across thresholds). Overall, the HSE set exhibits a significantlystronger network coexpression pattern than all of the alternativegene sets tested (with P < 0.05 for 99.9%, or 1,457 of 1,458 sets).Although significantly smaller than HSE, 2 of the 1,454 MSigDB

sets had relatively large connected components (Fig. S4C). Elevenother MSigDB sets appeared to have large connected compo-nents, but a single gene common to all 13 MSigDB sets, CARTPT(cocaine and amphetamine related transcript, prepropeptide),likely drives this result (Discussion and Fig. S4B).

Gene-Expression Covariance Patterns Are Consistent Across Individuals.Fig. 2 shows representative individual subject correlation matricesusing the HSE set. Brain samples are arranged in the same orderas in the group-averaged data in Fig. 1. Each individual has adifferent matrix size reflecting the number of brain samplesavailable from each network. The resulting matrices consistentlyshow a “blocked” pattern, whereby strong positive correlationsexist between the visual and somato/motor networks, as well as tothe auditory cortex. Strong negative correlations exist betweenthese sensory/motor regions and paralimbic and certain associa-tion networks, particularly the ventral attention and defaultnetwork regions.

Supragranular Molecular Profiles Are More Similar Within than AcrossNetworks. Gene-expression profiles are similar both within andbetween cortical networks for the HSE set. For example, regionswithin visual networks have similar molecular profiles to eachother and also to regions in the somato/motor networks andauditory cortex. Regions within paralimbic networks have similarexpression profiles to each other, as well as to regions in defaultand ventral attention networks. Within-network correlations areconsistently higher than out-of-network correlations (Fig. 3).Association networks have the lowest average difference in

within- and out-of-network correlation strengths. This reflectsrelatively greater heterogeneity in transcriptional profiles in theregions assigned to these networks (Figs. 1A and 3). The 17-network fcMRI parcellation identifies multiple subnetworkswithin the control network (31) that each has distributed com-ponents across prefrontal, parietal, and temporal associationcortices (e.g., orange and maroon networks in Fig. 1C). Regionswithin one subnetwork (Fig. 1C, orange) tend to have more similar

transcriptional expression to sensory/motor regions and the dorsalattention network, whereas another control subnetwork (Fig. 1C,maroon) tends to have more similar expression to paralimbic,ventral attention/salience, and default networks (Fig. 1A, seealso Fig. 7).Gene expression is generally expected to be highly consistent

across the cerebral cortex relative to other regions of the brain(4, 36, 37). Fig. 4 summarizes mean expression values for non-detrended and detrended (achieved by subtracting the meanexpression across all cortical brain samples for each probe) datafor the HSE set in one representative individual (H0351.1016).When nondetrended data are used, gene expression is highlysimilar across networks, consistent with expectations (4, 34). Thedetrended means for each gene (Fig. 4, Right) illustrate thedifferences in relative expression of these genes across networks.For example, the relative expression levels for these genes in theparalimbic network are an almost complete reversal of theirexpression in the visual network. Although the exact ordering ofthe 19 genes differs across individuals, the effect of detrendingwas consistent: KCNC3, COL6A1, SYT2, SCN4B, and VAMP1had the highest relative expression in the visual network acrossindividuals, whereas SCN3B, COL24A1, PRSS12, C1QL2, andTPBG had the lowest relative expression in the visual network.Across individuals this pattern was also consistently reversed forthe paralimbic and association networks.

Effect of Spatial Gradients on Molecular Profile Similarity. Gradedexpression along a principal axis is common in the adult cerebralcortex and often takes the form of rostrocaudal molecular gra-dients (8). To assess whether physical distance accounted fortranscriptional similarity, the MNI atlas coordinates of brain

Fig. 2. Individual subjects show consistent transcriptional profile groupings across networks. Individual correlation matrices for the HSE genes. Rows andcolumns are arranged by network, as in Fig. 1. Strong positive correlations along the diagonal indicate that transcriptional profiles are similar in brain regionsthat belong to the same network. Strong positive correlations on the off-diagonal indicate similarity between networks.

Fig. 3. Transcriptional profiles are more similar within-network than be-tween networks. (Left) HSE set transcriptional similarity (correlation) sum-marized by the seven-network parcellation. Regions that belong to the samenetwork have higher correlations than regions that belong to differentnetworks. Error bars show SEM across individuals. (Right) Surface represen-tation of plotted networks.

4 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1510903113 Krienen et al.

samples were sorted along the rostrocaudal axis for two rep-resentative individuals (Fig. 5). In each case, gene sets showed a“checkerboard” pattern, whereby positive correlations appearedboth on and off the diagonal. The extreme off-diagonal positivecorrelations indicate similarity in gene expression between cer-tain brain samples located far apart along the rostrocaudal axis.Negative correlations close to the diagonal indicate that regionscloser together in terms of rostrocaudal location have dissimilarmolecular profiles. Transcriptional variation of the HSE genes isnot captured by a simple rostrocaudal gradient.More generally, transcriptional similarity is expected to be

negatively correlated with spatial distance (34). Given that cor-ticocortical connectivity is densest at short distances (38), itcould be that spatial proximity is a primary driver of transcrip-tional similarity within and across networks for the HSE genes.Fig. 6 and Figs. S5–S7 plot the correlation strength in gene ex-pression between pairs of brain samples as a function of theEuclidean distance separating them for each individual, for HSEand four alternative gene sets. These results indicate that theinfluence of physical distance on transcriptional profile correla-tion is negative, consistent with expectations (r = −0.30 to −0.51for the HSE set). Beyond the expected negative spatial trend,color-coding the sample pairs reveals the additional influence ofnetwork identity (Fig. 6 and Figs. S5–S7). At equivalent spatialdistances, cross-network region pairs of visual, somato/motor, orauditory cortical regions have correlations that fall above thetrend line, whereas region pairs between the sensory and asso-ciation or paralimbic cortex tend to cluster below the trend line.Cross-network correlations were directly compared by binning

pairs of samples across five equally spaced Euclidean distancebins. At equivalent distances, transcriptional similarity was con-sistently higher among visual, somato/motor, and auditory cor-tical pairs than between those regions and certain associationnetwork regions (default, ventral attention) or paralimbic re-gions (e.g., red versus magenta points in Fig. S5) (two-sample ttests, P’s < 0.001 across individuals, family-wise error rate-cor-rected). Exceptions were the closest and most extreme distancebins, where there were insufficient numbers of pairs in eachcategory for most individuals. Comparison of sensory-associationor sensory-paralimbic pairs to association-paralimbic pairs atequivalent distance bins was also significant (magenta vs. bluepoints in Fig. S5) (two-sample t tests, P’s < 0.001 across indi-viduals, family-wise error rate-corrected).This separation of sample pair correlations by cortical type or

network assignment was strongest in the HSE set (Fig. 6 and Fig.S5), with diminished separation in other gene sets (Figs. S6 andS7). These results indicate that spatial proximity does not uni-

formly affect the tendencies of brain samples to have similar ordissimilar transcriptional profiles: at equivalent spatial distances,paralimbic and certain associational networks are more likely tohave similar expression profiles to each other than to auditory,visual, or somato/motor profiles.

Degree of Connectivity Predicts Transcriptional Similarity.Unimodaland hierarchically organized sensory pathways tend to form thedensest connections to adjacent areas (16, 38–40). In contrast,heteromodal association areas tend to support longer-range con-nections and to participate in distributed systems (15, 41, 42).Neuroimaging approaches have similarly demonstrated that the

relative balance of local versus distant connectivity is heteroge-nous across the cerebral mantle in humans (13). Unimodal sensorycortical regions display preferentially local connectivity, whereashigher order heteromodal association regions distributed acrossthe cortex have preferentially distant connectivity. The analysis inSepulcre et al. (13) relied on fMRI-based functional connectivitymeasures with limited spatial resolution and is an indirect measureof anatomical connectivity. Notably, task interactions can shift thebalance of local to distant connectivity, indicating dynamic influ-ences on these estimates. However, the broad connectivity prin-ciples observed in that study converge with basic observationsfrom nonhuman primate tracing (14, 15).To test the assumption that relative differences in expression of

these HSE genes are associated with differences in corticocorticalconnectivity, Fig. 7 examines whether relative differences in thedegree of local and long-range functional connectivity providefurther insight into how the gene profiles group together.Fig. 7A shows the relative preference for distant versus local

connectivity plotted across the human cortical surface fromSepulcre et al. (13) (Supporting Information). The surface mapshows the direct subtraction between the degree of positive cor-relations outside of and within local neighborhoods of voxelsacross the brain. Warmer colors indicate regions that have pref-erentially long-distance coupling, whereas cool colors indicateregions with preferentially local coupling. Fig. 7B plots the group-averaged HSE set correlation matrix for 17 networks. Valuesalong the rows indicate the average distant–local score for eachnetwork. Networks with similar connectional profiles also tend tohave similar gene-expression profiles. Fig. 7C shows the corre-lation between group-averaged HSE gene profiles for all corticalcomponents (Fig. 1A) and the absolute difference in their dis-tant–local degree connectivity score. The negative correlation(ρ = −0.38; P < 0.001) indicates that transcriptional profilesimilarity tends to decrease with more divergent distant–localdegree connectivity between pairs of brain regions.

Fig. 4. Relative differences in transcriptional profiles across networks arerevealed by detrending expression values. (Left) Mean expression values foreach of the 19 genes in the HSE gene set, plotted for seven networks. (Right)After mean expression is subtracted for each probe across cortical samples,relative differences in expression become evident across the seven networks.

Fig. 5. Transcriptional profiles can be similar even when located far aparton the cortical mantle. Two individual matrices from Fig. 2 are rearrangedaccording to rostrocaudal position on the cortical mantle. In each case,strong off-diagonal correlations indicate that regions spaced far apart havesimilar transcriptional profiles.

Krienen et al. PNAS Early Edition | 5 of 10

NEU

ROSC

IENCE

PNASPL

US

DiscussionOur analyses demonstrate that transcriptional expression of cer-tain genes reflects the classic subdivision of the cerebral cortexinto different types (sensory/motor, paralimbic, and heteromodalassociation). The genes that best distinguish these subdivisions—the HSE set—are selectively enriched in upper layers of cortexin humans relative to mouse (18).The shift of these genes to supragranular layers may be involved

in the evolution of increased long-range intracortical projectionsin primates generally, or humans specifically. The in situ hybrid-ization analysis in Zeng et al. (18) was limited to two cortical re-gions: visual cortex (BA17 and BA18) and temporal associationcortex (BA21, parts of BA22 and BA20). Because of the limitedsampling in that report, it was unknown whether the expression ofthese genes covaries more generally across the human cerebralcortex in a manner consistent with intracortical connectivity pro-files. Here we provide support for this hypothesis by showing thatthe expression profiles of these genes dissociate cortical regionswith predominately local connectivity (sensory/motor networks)from those that have a higher proportion of long-range connec-tivity in association and paralimbic cortical zones (Figs. 1–3 and 7).

Transcriptional Similarity Within and Across Networks. Transcrip-tional profiles were more similar within-network than out-of-network boundaries (Figs. 1C and 3 and Fig. S2). In many cases,cross-network transcriptional similarity follows expectations from

anatomic connectivity derived from nonhuman primate tracttracing (e.g., Fig. S2). Network groupings, such as those producedfrom fcMRI estimates, may be best conceived as a means foridentifying regions of the cortex with similar connectional prop-erties. The transcriptional similarity between these regions, then,could reflect those shared properties regardless of whether theyare directly coupled. For example, note that functional connectivityestimates do not group the default network with the ventral at-tention/salience network (Fig. 1B), yet their transcriptional profilesfor the HSE genes are similar (Figs. 1A and 6). This result is areminder that transcriptional similarity may primarily be indexingbroad classes of cortex and not necessarily markers of individualnetworks. In addition to their connectivity differences, these re-gions also tend to show protracted development, have distinctmetabolic profiles, and are more variable between individuals (11).Further subdivisions within the broad classes discussed here (sen-

sory/motor, paralimbic, association) could reveal additional structureof how transcriptional profiles vary across the cerebral cortex. Forexample, transcriptional profiles of the HSE genes are not uniformacross all of the association cortex: in certain cases neighboring re-gions of the prefrontal cortex have different transcriptional profiles,as do neighboring regions in the parietal and temporal associationcortices. The present results show that grouping by network isuseful for revealing where these transitions occur, and whether theytend to occur in concert across distributed, interconnected regions.The profiles of regions in the dorsal attention network take

transitional forms between the sensory/motor profiles and asso-

A B C

Fig. 7. Balance of local and distant coupling predicts transcriptional similarity. (A) Relative difference in distant–local degree connectivity measured byproportion of correlations that are within the local neighborhood versus distant correlations in resting state functional connectivity data [adapted fromSepulcre et al. (13)]. (B) Average distance score for 17 networks, listed along the group-averaged transcriptional similarity matrix for the HSE gene set.(C) Scatter plot showing the relationship between transcriptional similarity and degree connectivity score (from A) for all network pairs.

Fig. 6. HSE gene set transcriptional profiles are a function of spatial proximity as well as network identity and cortical type. Transcriptional profile corre-lations are plotted against Euclidean distance measurements for pairs of brain samples. Dark gray points are within-network pairs. Note that dark gray pointstend to group at the top of the graph, meaning they have higher correlations even at long distances. Red points are somato/motor network to visual networkpairs. Note that they tend to have high correlations despite long distances. Blue points are paralimbic or ventral attention to default network pairs. Note thatthey tend to have high correlations at both close and long distances. Magenta points are paralimbic or ventral attention to visual network or somato/motornetwork pairs. Note that they tend to have low correlations at all distances. Light gray points are all other combinations. Note that they tend to follow theoverall negative correlation and are particularly evident at the long-range, low-correlation corner of the graph.

6 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1510903113 Krienen et al.

ciation cortex profiles (Fig. 3). The putative homolog of thedorsal attention network in nonhuman primates forms a ca-nonical pathway linking the primary visual cortex to extrastriatevisual areas, the middle temporal area complex (MT+), posteriorparietal cortex, and the frontal eye fields (43, 44). Recent ana-tomical tracing in mice suggest that the visual processing path-way may be similarly organized into distributed, dorsal, andventral processing streams in that species (45). This findingwould suggest that it is a conserved pathway that may have beenpresent in the rodent-primate common ancestor. The dorsal at-tention network is a distributed network involved in sensory-motor integration and may represent an early, evolved prototypenetwork that has been expanded and elaborated in the distinctcortical networks that are observed in primates.The present results show that regions in the dorsal attention

network have transcriptional profiles that align most closely withgene-expression profiles of sensory/motor regions, but are alsosimilar to expression profiles of some association regions. Inparticular, dorsal attention network regions have similar tran-scriptional expression of the HSE genes to regions that partici-pate in systems involved in executive control (control network)(Figs. 1A and 6) (46, 47). Consistent with this pattern, humanfunctional connectivity analyses show that the dorsal attentionnetwork couples to both extrastriate visual regions, as well as toexecutive control regions distributed across prefrontal, parietal,and occipito-temporal cortices (Fig. 1) (48).In other instances, transcriptional agreements between regions

belonging to different networks signify an important departure fromthe pattern observed from fcMRI estimates. The most striking de-parture is the similarity in expression for the HSE genes betweenregions belonging to visual networks and those belonging to thesomato/motor networks and auditory cortex (Figs. 1, 2, 4, and 7).Here, gene expression is similar despite the lack of strong couplingbetween these networks indicated by human fcMRI estimates (Fig.1B), lack of direct strong anatomic connectivity in nonhuman pri-mates (49, 50), and dissimilarity in cytoarchitecture. Transcriptionalsimilarity of supragranular genes in sensory and motor zones mayreflect common tendencies of these zones to preferentially supportlocal connectivity (Fig. 7). Thus, a reasonable speculation is thatthese transcriptional profiles are marking broadly distinct types ofcortical architecture, which tend to be shared within a network, butare also shared across networks when the properties of the distinctnetworks are similar, such as may be the case for sensory and motorzones that are distributed across the cortex. Sensory and motorzones may be similar in the sense that they have preferential localcorticocortical connectivity, or in the sense that they lack thedistinct alternative properties of association networks. Either formof similarity—shared properties or shared absence of properties—could lead to their transcriptional profiles being similar relative toother cortical zones.

Transcriptional Variation Beyond Spatial Gradients. Gradients in theexpression of key transcription factors are strong drivers ofcortical arealization and patterning during development (51–53).In adults, gene expression is fairly uniform across cortical regions(8, 18, 34). However, in adult macaque monkeys, a number ofgenes with maximal variation in expression across cortical areasare expressed in rostrocaudal gradients. This finding may reflectpersistent traces of gradients in developmental origin and timingduring cortical neurogenesis (8).Although spatial proximity is likely to be a primary factor

constraining gene expression across the cerebral cortex (54, 55),the current analysis presents a different perspective: relativeexpression patterns of a small number of genes that are uniquelyexpressed in human upper cortical layers compared with mouseare correlated in distributed patterns across the cerebral cortex.Their spatial distributions across the cortex are closely linked tocortical subtypes and their associated connectivity profiles. Al-

though the dissociations were strongest for the HSE genes, othergene sets often showed weaker variants of the same phenome-non: in general, sensory and motor regions have more similargene profiles to themselves and each other, and paralimbic andassociation regions have more similar expression profiles tothemselves and each other.

Functional Ontology of Human Supragranular Enriched Genes. Somegenes within the HSE set have known functional roles in theformation or maintenance of corticocortical connectivity. Forexample, neurofilament protein-heavy (NEFH) predominantlylocalizes to corticocortical projecting neurons (56). Laminardensity of this protein is higher in V1 and in extrastriate visualareas relative to prefrontal areas, particularly in layer III (57).NEFH is also present in layer III long-range corticocorticalprojection neurons in the association cortex, but less so in paralimbicregions (56). Consistent with these protein-expression patterns,transcriptional expression of NEFH was consistently higher inthe visual network and lower in the paralimbic network (Fig. 4).γ-Synuclein (SNCG) regulates lipid metabolism in the brain (58)and plays a role in synaptic plasticity and formation (59, 60) andneurofilament network integrity (61). Other genes have a less-obvious role in connectivity. μ-Crystallin (CRYM), a cytosolicthyroid hormone-binding protein, is one such unexpected candidate;its precise physiological role in the brain is not yet known. Inter-pretations of our results must therefore take into account theincomplete record of genetic functional ontologies.Across individuals, the genes most consistently up-regulated in

the visual and somato/motor networks include SYT2, SCN4B,COL6A1, and VAMP1. COL6A1, which encodes an extracellularmatrix protein of the collagen family, regulates myelination inthe peripheral nervous system (62). It is also one of a set of genesthat are up-regulated in humans relative to chimpanzees (6), butthe reason for its laminar distribution in the human cerebralcortex is presently unknown. The differential VAMP1 expressionin sensory and somato/motor regions, versus paralimbic and as-sociation cortices, is corroborated by laminar differences evidentin the in situ hybridization patterns: VAMP1 is ubiquitouslyexpressed across layers in primary visual cortex (BA 17), but isrestricted and enriched in layer III pyramidal neurons in higher-order areas, including the temporal and dorsolateral prefrontalcortex (human.brain-map.org).Genes with highest expression in the association and paralimbic

cortices relative to sensory and motor regions tended to beconsistent across individuals. In particular, COL24A1, SCN3B,PRSS12, and C1QL2 consistently rank lowest in relative geneexpression in the visual network, and highest in the paralimbic anddefault network regions. The function of these genes within thecontext of supragranular laminar circuitry has not been wellcharacterized in model systems, because by definition they are notprominently expressed in upper layers in mouse. However, theC1QL family is thought to play a critical role in synapse formationand maintenance of synapses between climbing fibers from theinferior olive and Purkinje cells in the cerebellum (63), as well asin postnatal synapse elimination between retinal ganglion cells andthe lateral geniculate nucleus (64). It is also expressed in dentategyrus granule cells (65). The inclusion of several genes implicatedin sodium, calcium, and potassium ion channels suggests that thelaminar remodeling in the human cerebral cortex compared withthe mouse may have been accompanied by changes to the balanceof excitability within local and distributed cortical circuits.

Alternative Gene Sets.Overall, the HSE gene set has a significantlystronger association to corticocortical network connectivity prop-erties than 99.9% of the alternative gene sets (1,457 of 1,458)tested. Beyond direct comparisons to the HSE set, the MSigDBcollection presented an opportunity to discover other gene setswith potentially meaningful correlation structure in the context of

Krienen et al. PNAS Early Edition | 7 of 10

NEU

ROSC

IENCE

PNASPL

US

distributed connectivity networks. Two MSigDB sets, Neuro-transmitter Secretion and Regulated Secretory Pathway, retainedhigh correlation structure even when CARTPT—an outlier inother MSigDB sets—was omitted (Fig. S4). The NeurotransmitterSecretion set is a subset of 13 of 15 genes in the Regulated Se-cretory Pathway set. There is no overlap between these sets andthe HSE gene set; we consider these genes interesting candidatesfor further cross-species cerebral cortical characterization.Intriguingly, CARTPT appears to be selectively expressed by

human upper layer cortical pyramidal neurons, and is expressed inthe prefrontal, temporal association, as well as visual cortex. Inmouse, Cartpt is also primarily expressed in layer III in the cerebralcortex, and appears particularly enriched in primary sensory areasin that species. Cartpt expression has been best characterized in thehypothalamus. Its function there appears to be related to leptinregulation and shaping the negative response to appetite (66). Thefunction of this gene in the cerebral cortex is not well understood.In addition to the MSigDB collection, we considered four addi-

tional curated gene sets. The enrichment of HSE genes in upperlayers likely involves multiple cell types beyond pyramidal projec-tion neurons. That transcriptional coexpression of conserved uppermarkers (i.e., the Conserved Supergranular set) also, but to a lesserextent, dissociates network types is expected. Transcriptionalcovariation of genes forming the Human/Mouse Connectivity setwas recently shown in an elegant study (25) to dissociate fourhuman connectivity networks on the basis of a metric of higherwithin- than across-network association. Four of 19 of the HSEgenes (21%) were also contained in this set. Several genes in theRodent Connectivity and Conserved Supragranular sets are alsoshared with the Human/Mouse Connectivity set, which was gen-erally enriched for genes implicated in ion channel function andin synapses.The enrichment of these genes in human corresponds with

structural connectivity in homologous regions in mouse (25). Thus,the Human/Mouse Connectivity set may preferentially index genesassociated with connectivity in cortical regions that are conservedacross species. Consistent with this, the four genes (SYT2, NEFH,SV2C, SCN4B) that overlap with the HSE set tend to be up-reg-ulated in the visual and somato/motor networks and down-regu-lated in paralimbic and association networks. The HSE set capturesthe main properties of the Human/Mouse Connectivity set, butadditionally reveals substantial cross-network transcriptionalsimilarity in humans that reflects network properties acrosssensory/motor, paralimbic, and association cortical classes.

ConclusionsThe present results show that variation in network identity andconnectivity across regions of the cerebral cortex are mirrored bydifferent transcriptional profiles. For the genes analyzed in thisstudy, in particular those enriched in human upper layers, rela-tive variation in profiles was not exclusively driven by spatialproximity. Cortical subtype (sensory/motor, paralimbic, association)and associated corticocortical connectivity profiles align closely withcorrelations of transcriptional profiles. Molecular innovations ofupper-layer cortical architecture may be an important componentin the evolution of increased long-range corticocortical projectionsor other properties linked to these association areas (e.g., pro-tracted development). This theory does not imply that the expres-sion of these genes evolved to directly produce the particularforms of corticocortical connectivity observed across distributedassociation networks. Activity-dependent sculpting likely plays asubstantial role in the establishment and refinement of long-dis-tance connectivity, particularly in postnatal stages as associationcircuits mature (67, 68). Finally, we focused here on observationsthat tended to be consistent across the six postmortem brains (seealso ref. 69), but individual differences in genotype, gene expres-sion, and large-scale connectivity properties in larger cohortspromises to yield additional insight (25, 70).

Although functional classification of the HSE genes is in-complete with respect to their roles in circuit organization, theirtranscriptional profiles identify candidates for future functionalexperiments that elucidate their possible roles in the develop-ment or maintenance of corticocortical connectivity. Whethersupragranular enrichment of the genes analyzed here is trulyunique to humans, or common to other great apes or primatesbroadly, remains a hypothesis to be tested in future work.

MethodsMicroarray Datasets. Publicly available gene-expression datasets for humanpostmortem brains (n = 6) were obtained from the Allen Institute for BrainScience. Each individual’s dataset includes normalized microarray expressiondata for 58,692 probes measured from dissected samples from each post-mortem brain. Microarray data were filtered to retain samples from the cere-bral cortex (n = 4 left hemisphere only, n = 2 both left and right hemispheres).Before dissection, each postmortem brain was MRI-scanned, and T1-weightedvolumes were anatomically segmented using FreeSurfer’s recon-all pipeline(freesurfer.net). Individual subject anatomical volumes were registered to theMNI coordinate space. For more details on microarray and structural imagingdatasets, refer to the Allen Human Brain Atlas technical white papers (help.brain-map.org/display/humanbrain/Documentation).

Gene Probe Sets. Gene sets were obtained from published lists. A primarysource consisted of 1,454 annotated gene sets that belong to the MolecularSignatures Databasemaintained by the Broad Institute. MSigDB gene sets arenamed by their Gene Ontology term and consist of genes annotated by thatterm. The sets fall into three broad categories: Biological Process (a recog-nized series of events or molecular functions), Cellular Component (de-scribing locations at the level of subcellular or macromolecular complexes),and Molecular Function (the functions of a gene product). Four additionalalternative gene sets were analyzed: (i) Conserved Supragranular set, n = 14genes (18), (ii) Human Cortically Enriched set, n = 20 genes (2), (iii) RodentConnectivity set, n = 381 genes (24), and (iv) Human/Mouse Connectivity set,n = 136 genes (25). Gene information for MSigDB sets can be found online(software.broadinstitute.org/gsea/msigdb/index.jsp). The remaining four com-parison sets are available in Dataset S1.

Functional Connectivity Datasets. Consensus maps of 7 and 17 functionalconnectivity networks from resting-state data from 1,000 healthy youngadults in the Brain Genomics Superstruct Project (71) were used in the presentstudy. Participants provided written informed consent in accordance withguidelines set by institutional review boards of Harvard University or Part-ners Healthcare. The Yeo et al. (31) maps are publicly available for down-load (https://surfer.nmr.mgh.harvard.edu/fswiki/CorticalParcellation_Yeo2011),and are highly similar to alternative estimates (72). The method for clusteringhas been described elsewhere (31). Briefly, for each participant, the Pearson’sproduct moment correlation was computed between each surface vertex (n =18,715) and 1,175 uniformly distributed cortical regions of interest (ROIs). The“connectivity profile” of each surface vertex was its functional coupling tothese ROIs. Each participant’s 18,715 × 1,175 matrix of correlations wasbinarized to retain the top 10% of correlations before summing across subjectsto obtain an overall group estimate P. Therefore, the ith row and jth columnof thematrix Pwas the number of subjects whose correlations between the ithvertex and jth ROI are within the top 10% of correlations (within each indi-vidual subject). In other words, each matrix component took on integer valuesfrom 0 to 1,000. The connectivity profiles were clustered using a mixture ofvon Mises–Fisher distributions. For more details, see refs. 31 and 73. Becauseour previous analyses (31) identified solutions with 7 and 17 network clustersto be particularly stable, these were adopted for the present study.

Network Assignments. A volumetric 17-network parcellation in MNI space fromYeo et al. (31) was split into sets of components according to their locations andextent on the cortical surface. Specifically, a set of 114 cortical regions composedof roughly symmetric territories in the left and right hemispheres were definedby selecting vertices on a spherical inflation of each consensus map with respectto network boundaries, sulcal patterns, and confidence maps. “Confidencemaps” refer to estimates of instability of network assignments and suggest ei-ther uncertainty about proper network assignment in a given region, or antic-ipate further subdivisions of networks (31). A similar approach to obtain sets ofsubregions within cortical networks has been described elsewhere (74–76).

The cerebral cortical brain samples were assigned to network componentsby associating the MNI coordinate of each sample with the correspondingnetwork label at that coordinate. Brain samples that fell on the borders

8 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1510903113 Krienen et al.

between network components—or whose neighboring voxels did not havethe same network assignment—were excluded. Most of the excluded re-gions fell in subcortical structures (thalamus, basal ganglia) that were givencortical slab labels in the Allen Institute ontology (see Fig. S1 for locations ofincluded and excluded samples). Filtering subcortical structures was neces-sary as differences between cortical and subcortical transcriptional expres-sion dominate those within the cerebral cortex (34).

For visualization, the components were divided into seven groups (default,control, paralimbic, salience/ventral attention, dorsal attention, somato/motor,visual), which broadly correspond tomajor networks discussed in the literature.The paralimbic network includes paralimbic cortical regions, such as the medialsurface of the temporal lobe, orbitofrontal cortex, and subgenual cingulatecortex. Note that the somato/motor network clusters both pre- and postcentralgyrus, and also likely includes the loci of primary auditory core and belt areas inthe temporal lobe (31). Although strong anatomic connections betweenmotorand somatosensory regions are expected (77), we could not resolve from thefcMRI estimates whether the auditory cortex is truly functionally coupled tothese zones using fcMRI approaches (31). However, the transcriptional profiledata are able to make this distinction, so transcription data are separated bywhether they likely were obtained from motor (precentral gyrus) or primarysomatosensory areas (postcentral gyrus). Similarly, the close proximity of so-matosensory regions, particularly S2, to the auditory cortex and spatial limi-tations inherent to fMRI-based approaches induces clustering of auditoryregions to somato/motor regions because of partial voluming of the fcMRIdata (see figure 14 of ref. 31). Transcriptional profiles within auditory regions,which do not suffer from this spatial blurring, are assessed separately from thesomato/motor network in the present results.

Molecular Profile Correlation Analysis. When multiple probes in the micro-array data existed for a given gene, expression values were averaged acrossprobes. For each gene set, and for each postmortem brain, pairwise corre-lations of expression values of the probes were computed for all possible pairsof network components.

Because gene profiles are in general highly similar across the cerebralcortex (4), expression values were detrended by subtracting the mean ex-pression value across samples before correlation analyses. Because the pres-ence of outliers in a gene set can potentially inflate or deflate the correlationcoefficient, rank-signed Spearman’s ρ is used to generate correlation coeffi-cients for each gene set (see Fig. S4A for a comparison between Pearson’sproduct-moment correlation and Spearman’s ρ). Rows and columns of theresulting cross-correlation matrices were arranged so that components thatbelong to the same functional connectivity network are grouped together. Forbrains in which data from both hemispheres were available (H0351.2001 andH0351.2002), left hemisphere and right hemisphere components were in-terleaved within each network. For visualization, correlations were thresh-olded at ρ = ± 0.1.

To obtain a group summary of molecular profile agreements within andacross networks, the correlations for all regions belonging to the samenetwork (excluding self-correlations at the diagonal) were averaged. Thesewere averaged across individuals. Sixteen of the 114 regions did not containbrain samples from any of the individuals and were excluded from analysis,leaving 98 regions.

To determine whether correlations of gene expression between pairs ofthe 17 networks are significantly different from zero, average expression of

each gene was computed for each network by averaging across all brainsamples that fell within a given network of the 17-network parcellation (Fig.S3A). The labels of the HSE genes corresponding to the averaged expressionvalues were permuted independently for each network (retaining the samepermutation order across subjects for a given permutation). The resultingcorrelation matrix was computed to construct the null distribution (10,000permutations). This procedure breaks the correlation between samples whileretaining all other data structure. Correlations surviving an FDR-correctedvalue of q < 0.05 are reported. See Supporting Information for additional details.

To compare gene sets, a null distribution was obtained by randomly ex-changing the 19 HSE genes with those of the alternative set and computingthe statistic of interest using the permuted gene sets (Fig. S3B). For example,for the Rodent Connectivity set, the 19 HSE genes were pooled with the 381Rodent Connectivity genes. The pool was then randomly and repeatedly(10,000 permutations) split into two sets of 19 and 381 genes, and the ab-solute difference between the correlation matrices of the two gene sets wascomputed. The resulting “difference matrix” was compared with the dif-ference matrix computed from the two original gene sets. Only correlationssurviving an FDR-corrected value of q < 0.05 are interpreted. See SupportingInformation for additional details.

In addition to comparing individual elements in the correlationmatrices oftwo gene sets, we also used a global metric called network-based statistic (35)to examine whether there is significant correlation structure in a connec-tivity matrix of interest. Briefly, NBS is a nonparametric statistical test used toisolate components (sets of vertices connected by suprathreshold edges) ofan n × n (here 98 × 98, where each element is a network component fromthe 17 network parcellation) undirected connectivity matrix that differs be-tween two populations (in the present case, between two gene sets). NBS is aprocedure to control family-wise error rate in the weak sense. Specifically, thedifference matrix between the HSE and a given alternative gene set is thresh-olded, and the largest connected component is computed. This process is re-peated for the permuted difference matrices created by the pooling andrandom splitting procedure described above, and component sizes between theoriginal difference matrix and the permuted difference matrix are compared.

ACKNOWLEDGMENTS. We thank Steven McCarroll and Arpiar Saunders forhelpful discussions. Data were provided (in part) by the Brain GenomicsSuperstruct Project of Harvard University and Massachusetts General Hospital,(Principal Investigators: R.L.B., Joshua Roffman, and Jordan Smoller), withsupport from the Center for Brain Science Neuroinformatics Research Group,the Athinoula A. Martinos Center for Biomedical Imaging, and the Center forHuman Genetics Research. This work was supported by a George WashingtonUniversity Centers and Institute’s Facilitating Fund Intramural Award to theMind-Brain Institute; National University of Singapore Tier 1, Singapore MOETier 2 (MOE2014-T2-2-016); National University of Singapore Strategic Re-search Grant DPRT/944/09/14; National University of Singapore School of Med-icine Grant R185000271720; Singapore National Medical Research CouncilCBRG14nov007, NMRC/CG/013/2013, National University of Singapore YoungInvestigator Award; the James S. McDonnell Foundation (220020293); and aMassachusetts General Hospital Executive Committee on Research TostesonPostdoctoral Fellowship award (to T.G.). This research also used resources pro-vided by the Center for Functional Neuroimaging Technologies (P41EB015896),and instruments supported by Grants 1S10RR023401, 1S10RR019307, and1S10RR023043 from the Athinoula A. Martinos Center for Biomedical Imagingat the Massachusetts General Hospital.

1. Khaitovich P, Enard W, Lachmann M, Pääbo S (2006) Evolution of primate gene ex-

pression. Nat Rev Genet 7(9):693–702.2. Konopka G, et al. (2012) Human-specific transcriptional networks in the brain. Neuron

75(4):601–617.3. Oldham MC, et al. (2008) Functional organization of the transcriptome in human

brain. Nat Neurosci 11(11):1271–1282.4. Lein ES, et al. (2007) Genome-wide atlas of gene expression in the adult mouse brain.

Nature 445(7124):168–176.5. Babbitt CC, et al. (2010) Both noncoding and protein-coding RNAs contribute to gene

expression evolution in the primate brain. Genome Biol Evol 2(0):67–79.6. Cáceres M, et al. (2003) Elevated gene expression levels distinguish human from non-

human primate brains. Proc Natl Acad Sci USA 100(22):13030–13035.7. Zapala MA, et al. (2005) Adult mouse brain gene expression patterns bear an em-

bryologic imprint. Proc Natl Acad Sci USA 102(29):10357–10362.8. Bernard A, et al. (2012) Transcriptional architecture of the primate neocortex. Neuron

73(6):1083–1099.9. Sherwood CC, Duka T (2012) Now that we’ve got the map, where are we going?

Moving from gene candidate lists to function in studies of brain evolution. Brain

Behav Evol 80(3):167–169.10. Hill J, et al. (2010) Similar patterns of cortical expansion during human development

and evolution. Proc Natl Acad Sci USA 107(29):13135–13140.

11. Buckner RL, Krienen FM (2013) The evolution of distributed association networks inthe human brain. Trends Cogn Sci 17(12):648–665.

12. Kaas JH (1997) Topographic maps are fundamental to sensory processing. Brain ResBull 44(2):107–112.

13. Sepulcre J, et al. (2010) The organization of local and distant functional connectivityin the human brain. PLOS Comput Biol 6(6):e1000808.

14. Markov NT, et al. (2013) The role of long-range connections on the specificity of themacaque interareal cortical network. Proc Natl Acad Sci USA 110(13):5187–5192.

15. Markov NT, et al. (2013) Cortical high-density counterstream architectures. Science342(6158):1238406.

16. Kaas J (2002) Cortical areas and patterns of cortico-cortical connections. Cortical Areas:Unity and Diversity, eds Schütz A, Miller R (CRC Press, Boca Raton, FL), pp 179–191.

17. Sherwood CC, Subiaul F, Zawidzki TW (2008) A natural history of the human mind:Tracing evolutionary changes in brain and cognition. J Anat 212(4):426–454.

18. Zeng H, et al. (2012) Large-scale cellular-resolution gene profiling in human neo-cortex reveals species-specific molecular signatures. Cell 149(2):483–496.

19. Marín-Padilla M (1992) Ontogenesis of the pyramidal cell of the mammalian neo-cortex and developmental cytoarchitectonics: A unifying theory. J Comp Neurol321(2):223–240.

20. Hutsler JJ, Lee D-G, Porter KK (2005) Comparative analysis of cortical layering andsupragranular layer enlargement in rodent carnivore and primate species. Brain Res1052(1):71–81.

Krienen et al. PNAS Early Edition | 9 of 10

NEU

ROSC

IENCE

PNASPL

US

21. Elston GN, Benavides-Piccione R, DeFelipe J (2001) The pyramidal cell in cognition: Acomparative study in human and monkey. J Neurosci 21(17):RC163.

22. Petanjek Z, et al. (2011) Extraordinary neoteny of synaptic spines in the human pre-frontal cortex. Proc Natl Acad Sci USA 108(32):13281–13286.

23. Bianchi S, et al. (2013) Dendritic morphology of pyramidal neurons in the chimpanzeeneocortex: Regional specializations and comparison to humans. Cereb Cortex 23(10):2429–2436.

24. Wolf L, Goldberg C, Manor N, Sharan R, Ruppin E (2011) Gene expression in the ro-dent brain is associated with its regional connectivity. PLOS Comput Biol 7(5):e1002040.

25. Richiardi J, et al.; IMAGEN consortium (2015) BRAIN NETWORKS. Correlated geneexpression supports synchronous activity in brain networks. Science 348(6240):1241–1244.

26. Raichle ME (2011) The restless brain. Brain Connect 1(1):3–12.27. Buckner RL, Krienen FM, Yeo BTT (2013) Opportunities and limitations of intrinsic

functional connectivity MRI. Nat Neurosci 16(7):832–837.28. Vincent JL, et al. (2007) Intrinsic functional architecture in the anaesthetized monkey

brain. Nature 447(7140):83–86.29. Margulies DS, et al. (2009) Precuneus shares intrinsic functional architecture in hu-

mans and monkeys. Proc Natl Acad Sci USA 106(47):20069–20074.30. Margulies DS, Petrides M (2013) Distinct parietal and temporal connectivity profiles of

ventrolateral frontal areas involved in language production. J Neurosci 33(42):16846–16852.

31. Yeo BTT, et al. (2011) The organization of the human cerebral cortex estimated byintrinsic functional connectivity. J Neurophysiol 106(3):1125–1165.

32. Hutchison RM, et al. (2012) Functional connectivity of the frontal eye fields in humansand macaque monkeys investigated with resting-state fMRI. J Neurophysiol 107(9):2463–2474.

33. Choi EY, Yeo BTT, Buckner RL (2012) The organization of the human striatum esti-mated by intrinsic functional connectivity. J Neurophysiol 108(8):2242–2263.

34. Hawrylycz MJ, et al. (2012) An anatomically comprehensive atlas of the adult humanbrain transcriptome. Nature 489(7416):391–399.

35. Zalesky A, Fornito A, Bullmore ET (2010) Network-based statistic: identifying differ-ences in brain networks. Neuroimage 53(4):1197–1207.

36. Khaitovich P, et al. (2004) A neutral model of transcriptome evolution. PLoS Biol 2(5):682–689.

37. Yamamori T, Rockland KS (2006) Neocortical areas, layers, connections, and geneexpression. Neurosci Res 55(1):11–27.

38. Markov NT, et al. (2014) A weighted and directed interareal connectivity matrix formacaque cerebral cortex. Cereb Cortex 24(1):17–36.

39. Jones EG, Powell TP (1970) An anatomical study of converging sensory pathwayswithin the cerebral cortex of the monkey. Brain 93(4):793–820.

40. Pandya DN, Kuypers HG (1969) Cortico-cortical connections in the rhesus monkey.Brain Res 13(1):13–36.

41. Goldman-Rakic PS (1988) Topography of cognition: Parallel distributed networks inprimate association cortex. Annu Rev Neurosci 11(1):137–156.

42. Mesulam MM (1990) Large-scale neurocognitive networks and distributed processingfor attention, language, and memory. Ann Neurol 28(5):597–613.

43. Maunsell JH, van Essen DC (1983) The connections of the middle temporal visual area(MT) and their relationship to a cortical hierarchy in the macaque monkey. J Neurosci3(12):2563–2586.

44. Shadlen MN, NewsomeWT (2001) Neural basis of a perceptual decision in the parietalcortex (area LIP) of the rhesus monkey. J Neurophysiol 86(4):1916–1936.

45. Wang Q, Sporns O, Burkhalter A (2012) Network analysis of corticocortical connec-tions reveals ventral and dorsal processing streams in mouse visual cortex. J Neurosci32(13):4386–4399.

46. Dosenbach NUF, Fair DA, Cohen AL, Schlaggar BL, Petersen SE (2008) A dual-networksarchitecture of top-down control. Trends Cogn Sci 12(3):99–105.

47. Power JD, Petersen SE (2013) Control-related systems in the human brain. Curr OpinNeurobiol 23(2):223–228.

48. Vincent JL, Kahn I, Snyder AZ, Raichle ME, Buckner RL (2008) Evidence for a fronto-parietal control system revealed by intrinsic functional connectivity. J Neurophysiol100(6):3328–3342.

49. Kaas JH (1989) Why does the brain have so many visual areas? J Cogn Neurosci 1(2):121–135.

50. Felleman DJ, Van Essen DC (1991) Distributed hierarchical processing in the primatecerebral cortex. Cereb Cortex 1(1):1–47.

51. Rubenstein JL, Shimamura K, Martinez S, Puelles L (1998) Regionalization of theprosencephalic neural plate. Annu Rev Neurosci 21(1):445–477.

52. Fukuchi-Shimogori T, Grove EA (2001) Neocortex patterning by the secreted signalingmolecule FGF8. Science 294(5544):1071–1074.

53. O’Leary DD, Sahara S (2008) Genetic regulation of arealization of the neocortex. CurrOpin Neurobiol 18(1):90–100.

54. Chen C-H, et al. (2011) Genetic influences on cortical regionalization in the humanbrain. Neuron 72(4):537–544.

55. Chen C-H, et al. (2012) Hierarchical genetic organization of human cortical surfacearea. Science 335(6076):1634–1636.

56. Hof PR, Nimchinsky EA, Morrison JH (1995) Neurochemical phenotype of cortico-cortical connections in the macaque monkey: Quantitative analysis of a subset ofneurofilament protein-immunoreactive projection neurons in frontal, parietal, tem-poral, and cingulate cortices. J Comp Neurol 362(1):109–133.

57. Hof PR, Morrison JH (1995) Neurofilament protein defines regional patterns of cor-tical organization in the macaque monkey visual system: A quantitative immuno-histochemical analysis. J Comp Neurol 352(2):161–186.

58. Guschina I, et al. (2011) Lipid classes and fatty acid patterns are altered in the brain ofγ-synuclein null mutant mice. Lipids 46(2):121–130.

59. Clayton DF, George JM (1998) The synucleins: A family of proteins involved in synapticfunction, plasticity, neurodegeneration and disease. Trends Neurosci 21(6):249–254.

60. Greten-Harrison B, et al. (2010) αβγ-Synuclein triple knockout mice reveal age-dependent neuronal dysfunction. Proc Natl Acad Sci USA 107(45):19573–19578.

61. Surguchov A, Palazzo RE, Surgucheva I (2001) Gamma synuclein: Subcellular locali-zation in neuronal and non-neuronal cells and effect on signal transduction. CellMotil Cytoskeleton 49(4):218–228.

62. Chen P, Cescon M, Megighian A, Bonaldo P (2014) Collagen VI regulates peripheralnerve myelination and function. FASEB J 28(3):1145–1156.

63. Kakegawa W, et al. (2015) Anterograde C1ql1 signaling is required in order to de-termine and maintain a single-winner climbing fiber in the mouse cerebellum.Neuron 85(2):316–329.

64. Stevens B, et al. (2007) The classical complement cascade mediates CNS synapseelimination. Cell 131(6):1164–1178.

65. Iwano T, Masuda A, Kiyonari H, Enomoto H, Matsuzaki F (2012) Prox1 postmitoticallydefines dentate gyrus cells by specifying granule cell identity over CA3 pyramidal cellfate in the hippocampus. Development 139(16):3051–3062.

66. Rogge G, Jones D, Hubert GW, Lin Y, Kuhar MJ (2008) CART peptides: Regulators ofbody weight, reward and other functions. Nat Rev Neurosci 9(10):747–758.

67. Krubitzer L, Kaas J (2005) The evolution of the neocortex in mammals: How is phe-notypic diversity generated? Curr Opin Neurobiol 15(4):444–453.

68. Miller DJD, et al. (2012) Prolonged myelination in human neocortical evolution. ProcNatl Acad Sci USA 109(41):16480–16485.

69. Hawrylycz M, et al. (2015) Canonical genetic signatures of the adult human brain. NatNeurosci, 10.1038/nn.4171.

70. Wang G-Z, et al. (2015) Correspondence between resting-state activity and brain geneexpression. Neuron 88(4):659–666.

71. Holmes AJ, et al. (2015) Brain Genomics Superstruct Project initial data release withstructural, functional, and behavioral measures. Sci Data 2:150031.

72. Power JD, et al. (2011) Functional network organization of the human brain. Neuron72(4):665–678.

73. Lashkari D, Vul E, Kanwisher N, Golland P (2010) Discovering structure in the space offMRI selectivity profiles. Neuroimage 50(3):1085–1098.

74. Baker JT, et al. (2014) Disruption of cortical association networks in schizophrenia andpsychotic bipolar disorder. JAMA Psychiatry 71(2):109–118.

75. Krienen FM, Yeo BTT, Buckner RL (2014) Reconfigurable task-dependent functionalcoupling modes cluster around a core functional architecture. Philos Trans R Soc LondB Biol Sci 369(1653):20130526.

76. Yeo BTT, Tandi J, Chee MWL (2015) Functional connectivity during rested wakefulnesspredicts vulnerability to sleep deprivation. Neuroimage 111(C):147–158.

77. Kaas JH (2004) Evolution of somatosensory and motor cortex in primates. Anat Rec ADiscov Mol Cell Evol Biol 281(1):1148–1156.

10 of 10 | www.pnas.org/cgi/doi/10.1073/pnas.1510903113 Krienen et al.